Multimedia Compression: Techniques, Formats & Applications

Why Compression Matters

Multimedia files (images, audio, video) are inherently large. A single 1-minute HD video can exceed 1 GB uncompressed. Compression reduces file size while preserving perceptual quality, enabling:

• Faster streaming (e.g., YouTube, Netflix).
• Lower storage costs (e.g., cloud backups, mobile devices).
• Efficient transmission (e.g., 4G/5G data limits).

Key Trade-off: Compression vs. Quality

• Lossless: No data loss (e.g., ZIP files), but limited compression.
• Lossy: Sacrifices minor details for high compression (e.g., MP3, JPEG).
• Hybrid: Combines both (e.g., H.264 video codecs).

1. Lossless Compression: No Data Loss

Definition: Reduces file size without discarding any original data. Reconstructs the exact original file upon decompression.

How It Works

Lossless compression exploits redundancy in data:

• Repetition: Storing "aaaabbb" as "4a3b" (run-length encoding).
• Patterns: Using dictionaries (e.g., LZW in GIF/PNG).
• Entropy: Assigning shorter codes to frequent symbols (Huffman coding).

flowchart TD
    A["Original Data"] -->|"Redundancy Analysis"| B["Dictionary/Codebook"]
    B -->|"Replace Patterns"| C["Compressed Data"]
    C -->|"Decompress"| B
    B -->|"Reconstruct"| A

Real-World Example: eSewa’s Transaction Logs

• Problem: eSewa processes millions of transactions daily, each with repetitive fields (e.g., user ID, timestamp).
• Solution: Lossless compression (e.g., Zstandard) reduces database size by 60–80%, speeding up queries without losing transaction records.

Common Lossless Formats

2. Lossy Compression: Sacrificing Quality for Efficiency

Definition: Permanently removes "irrelevant" data (e.g., frequencies humans can’t hear, color imperceptible to the eye). Achieves 10x–100x smaller files than lossless.

How It Works

Lossy compression targets human perception limits:

• Audio (MP3): Removes frequencies >20 kHz (inaudible to most) and quantizes low-amplitude sounds.
• Images (JPEG): Discards high-frequency details (e.g., fine textures) using Discrete Cosine Transform (DCT).
• Video (H.264): Exploits temporal redundancy (similar frames) via motion compensation.

flowchart LR
    A["Original Frame"] --> B["DCT: Frequency Domain"]
    B --> C["Quantize: Discard High Frequencies"]
    C --> D["Entropy Coding (Huffman)"]
    D --> E["Compressed Frame"]

Real-World Example: YouTube’s Video Streaming

• Problem: A 1080p video requires ~10 Mbps uncompressed. Mobile users have limited bandwidth.
• Solution: YouTube uses H.264/AVC (lossy) to stream at ~2 Mbps, reducing file size by 80% while maintaining "good enough" quality for most viewers.

Common Lossy Formats

3. Hybrid Compression: Best of Both Worlds

Definition: Combines lossless and lossy techniques for optimal trade-offs. Used in modern codecs like H.265 (HEVC) and AV1.

How It Works

1. Lossy stage: Reduces data size aggressively (e.g., DCT for video).
2. Lossless stage: Compresses residuals (remaining errors) without loss.
3. Metadata: Stores parameters for perfect reconstruction.

mindmap
  root((Hybrid Compression))
    Lossy Stage
      DCT
      Quantization
    Lossless Stage
      Arithmetic Coding
      Context Modeling
    Metadata
      Bitstream Syntax

Real-World Example: Daraz’s Product Images

• Problem: Daraz hosts millions of product images (e.g., clothing, electronics). Storing all as lossless PNG would require terabytes of space.
• Solution: Uses WebP (hybrid: lossy for photos, lossless for logos) to reduce image sizes by 30–50% while keeping visual quality acceptable for online shopping.

Comparison: Lossless vs. Lossy vs. Hybrid

4. Compression Techniques Deep Dive

A. Entropy Coding (Huffman, Arithmetic)

• Idea: Assign shorter codes to frequent symbols.
• Example: In English text, "e" appears more often than "z". Huffman coding assigns 0 to "e" and 100 to "z".
• Real Use: Used in JPEG, MP3, and ZIP files.

B. Transform Coding (DCT, Wavelet)

• Idea: Convert data into a frequency domain where redundancy is easier to exploit.
  • DCT: Used in JPEG (separates image into "smooth" and "detail" frequencies).
  • Wavelet: Used in JPEG 2000 (better for medical/scientific images).

graph LR
    A["Original Image"] --> B["DCT: 8x8 Blocks"]
    B --> C["Quantize: Keep Low Frequencies"]
    C --> D["Zig-Zag Scan + Huffman"]

C. Predictive Coding (Delta Encoding)

• Idea: Store differences between frames (used in video).
• Example: In a talking-head video, only the mouth region changes. Predictive coding stores the first frame fully and subsequent frames as deltas (changes).

5. Audio Compression: MP3 vs. AAC vs. Opus

How Audio Compression Works

1. Filtering: Remove frequencies outside human hearing range (0–20 kHz).
2. Psychoacoustic Model: Mask loud sounds with quiet ones (e.g., a drum beat masks a soft guitar).
3. Quantization: Reduce precision of less important samples.

Comparison Table

Real-World Example: Ncell’s Voice Calls

• Problem: Uncompressed voice requires 64 kbps. Mobile networks have limited bandwidth.
• Solution: Uses Opus codec (adaptive bitrate) to reduce data usage by 70% while keeping call quality clear.

6. Video Compression: H.264 vs. H.265 vs. AV1

Key Techniques

1. Intra-frame compression: Compress each frame independently (like JPEG for video).
2. Inter-frame compression: Exploit similarities between frames (motion compensation).
3. Macroblocks: Divide frames into 16x16 pixel blocks for efficient encoding.

Comparison Table

7. Worked Example: Calculating JPEG Compression Ratio

Problem: A 24-bit RGB image (8 bits per channel) is 1024×768 pixels. After JPEG compression, it’s saved as 50 KB. What’s the compression ratio?

Solution:

1. Uncompressed size: $1024\times 768\times 3\text{ bytes}=2,359,296\text{ bytes}\approx 2.36\text{ MB}$.
2. Compressed size: 50 KB = 0.05 MB.
3. Compression ratio: $\frac{2.36}{0.05}=47.2:1$.

Real-World Tie-In: This ratio is typical for Facebook profile pictures. If every user uploaded uncompressed images, Facebook’s storage costs would skyrocket.

8. Challenges and Trade-offs

In the Real World

1. Khalti’s Transaction Data

   • Idea Used: Lossless compression (e.g., Snappy or Zstandard).
   • How: Khalti processes thousands of transactions/sec. Compressing logs reduces database load, ensuring low-latency responses for users.

2. Pathao’s Ride-Hailing App

   • Idea Used: Hybrid video compression (H.265 for driver-passenger calls).
   • How: Calls use adaptive bitrate streaming to work on 2G–5G networks, balancing quality and data usage.

3. NTC’s Fiber-Optic Backbone

   • Idea Used: Audio/video compression (AAC for VoIP, H.264 for video conferencing).
   • How: Compression reduces bandwidth needs, enabling more simultaneous calls on limited fiber capacity.

Exam Tip

What Examiners Look For

1. Definitions: Clearly distinguish lossless vs. lossy vs. hybrid compression.
2. Algorithms: Know Huffman coding, DCT, and motion compensation steps.
3. Formats: Match formats to use cases (e.g., FLAC for audio, WebP for web images).
4. Calculations: Practice compression ratio and bitrate problems.
5. Real-World Links: Relate concepts to Nepali apps (e.g., eSewa logs, Daraz images).

Common Pitfalls

• Confusing JPEG (lossy) with PNG (lossless).
• Ignoring perceptual models (e.g., psychoacoustics in MP3).
• Overlooking metadata in hybrid codecs (e.g., H.264’s SEI messages).

Sample Exam Question & Answer

Q: Explain how MP3 compression reduces file size without significantly affecting audio quality. A:

1. Frequency Filtering: Removes frequencies >20 kHz (inaudible to humans).
2. Psychoacoustic Model: Discards sounds masked by louder frequencies (e.g., bass masking high notes).
3. Quantization: Reduces precision of less important samples (e.g., quiet guitar strings).
4. Huffman Coding: Assigns shorter codes to frequent data patterns. Visual: Include a psychoacoustic masking curve diagram to show which frequencies are discarded.